An automatic text reader using neural networks

This paper proposes an Arabic typewritten text reader using neural networks. The idea is based on the way in which humans read. The system´s input is real newspaper texts written in the most common Arabic font (Naskh). The system predicts the size of the font, and uses it in separating lines, words and sub-words. Then, it scans the text to recognize its individual characters using a set of nine neural networks according to a certain procedure. The whole text is then rebuilt and stored to be used by any application. Using neural networks in segmentation results in an accurate and fast performance. Some enhancements are proposed in order to reach a more powerful and general version of this system